finds.dev← search

// the find

NVIDIA/DeepRecommender

★ 1,706 · Python · MIT · updated May 2021

Deep learning for recommender systems

A PyTorch implementation of the deep autoencoder approach for collaborative filtering described in NVIDIA's 2017 paper. It's a research artifact, not a production library — trained and benchmarked against the Netflix Prize dataset. If you want to understand how autoencoders can beat matrix factorization on rating prediction, this is a clean, readable reference.

The SELU activation + iterative output re-feeding trick is genuinely interesting and the paper results are reproducible from this code. Data pipeline handles the full Netflix Prize dataset without falling over. Azkaban job definitions show how NVIDIA actually scheduled these experiments internally. Tensorboard integration is wired in from the start, not bolted on.

Dead since 2021 and requires Python 3.6 and CUDA 8 — both ancient, and PyTorch has broken backwards compatibility multiple times since. Netflix Prize data is no longer freely downloadable from the official source, so the primary benchmark is hard to reproduce. No serving or inference API — you get a predictions file, nothing you can slot into a real system. Only supports explicit rating data; implicit feedback (clicks, views) is not handled, which is what most real recommenders actually need.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →